www.nature.

com/scientificreports

OPEN The tight-binding formulation of

the Kronig-Penney model
F. Marsiglio & R. L. Pavelich

Electronic band structure calculations are frequently parametrized in tight-binding form; the latter
Received: 3 October 2017
representation is then often used to study electron correlations. In this paper we provide a derivation
Accepted: 23 November 2017 of the tight-binding model that emerges from the exact solution of a particle bound in a periodic one-
Published: xx xx xxxx dimensional array of square well potentials. We derive the dispersion for such a model, and show that
an effective next-nearest-neighbour hopping parameter is required for an accurate description. An
electron-hole asymmetry is prevalent except in the extreme tight-binding limit, and emerges through a
“next-nearest-neighbour” hopping term in the dispersion. We argue that this does not necessarily imply
next-nearest-neighbour tunneling; this assertion is demonstrated by deriving the transition amplitudes
for a two-state effective model that describes a double-well potential, which is a simplified precursor
to the problem of a periodic array of potential wells. A next-nearest-neighbour tunneling parameter is
required for an accurate description even though there are no such neighbours.

The notion of an “effective model” or “effective potential” pervades essentially all of physics. Even at the under-
graduate level, for example, it is worth emphasizing that the lowly harmonic oscillator potential is really merely
an “effective potential”. In reality all potentials are generally more complicated—the spring will eventually stretch
inelastically—and any potential has a useful domain of applicability in every problem.
One can go further with “effective models”, with perhaps the best-known example being Feynman’s descrip-
tion of the ammonia molecule as a two-state system1. This sort of description is worthwhile for certain aspects
of the problem, such as the time dependence of the wave function, but it often remains unclear how parameters
required in the effective model are related to underlying “microscopic” characteristics of the same problem. A
more concrete realization of the Feynman two-state system is the double-well potential. In ref.2 (see also ref.3)
the states describing a particle in such a system were determined from the basic parameters, namely the barrier
height and width. At the same time, a “toy model” describing a two-state system with a single parameter, a transi-
tion amplitude t, to describe tunneling through the barrier (analogous to the amplitude for the nitrogen atom to
tunnel from below to above the plane of hydrogen atoms in the ammonia example), was shown to very accurately
describe the ground state splitting calculated by solving the complete Schrödinger equation. This was an example
of a case where the original model, with a Hilbert space consisting of an infinite number of states was approxi-
mately mapped onto a “toy” or “effective” model consisting of just two states.
In this paper we want to focus on a natural extension of this model, which occurs often in condensed matter,
and leads to a “starting” model for studying strongly correlated electron systems. In this case we start with a peri-
odic array of some potential that gives rise to energy bands, whose characteristics require a complete solution to
the Schrödinger equation. This calculation in principle involves an infinite Hilbert space, in two senses. First, even
a single potential well, representing a single atom with which an electron interacts, requires an infinite Hilbert
space. However, for a solid there are a large number of these wells—an infinite number if we allow the solid to
go on forever. This latter infinity is handled analytically through Bloch’s theorem4–7, which allows solution of the
electron wave function in the infinite periodic array in terms of the solution within a single unit cell (see also ref.8
where some context for this theorem is provided). Even with Bloch’s theorem, however, an infinite Hilbert space
is required to describe the (infinite) set of energy bands that emerges from the periodicity. An effective model,
known as a tight-binding model, reduces the infinite Hilbert space down to an N-dimensional Hilbert space,
where N is the number of atoms. This description is entirely analogous to the reduction of the double-well poten-
tial to a 2-dimensional Hilbert space, involving a single tunneling parameter t, and in fact, even as N → ∞, only a
single parameter t, called the “tunneling amplitude”, remains.
This “down-folding” process is often carried out for complete descriptions of rather complicated systems,
often with approximations or parameterizations along the way9. The purpose of this paper is to implement this

Department of Physics, University of Alberta, Edmonton, AB, T6G 2E1, Canada. Correspondence and requests for
materials should be addressed to F.M. (email: [email protected]) or R.L.P. (email: [email protected])

Scientific REPOrTS | 7: 17041 | DOI:10.1038/s41598-017-17223-2 1

www.nature.com/scientificreports/

Figure 1. A pictorial representation of the periodic potential in the Kronig-Penney model, illustrating wells of
depth V0 and width w separated from one another by barriers of width b.

procedure using the simplest model possible, i.e. the one-dimensional Kronig-Penney model10. The result is a
derivation of the one or two parameters in the “effective” tight-binding model, in terms of the microscopic param-
eters that describe the original Kronig-Penney model. This calculation is possible because some aspects of the
one-dimensional Kronig-Penney model are known analytically.
For reasons that will become clear as we proceed, comparisons are required with the double-well case (i.e. a
Kronig-Penney model with only 2 cells). This solution, performed for a configuration pertinent to the infinite-cell
Kronig-Penney model, is described in Appendix A. Our numerical solutions confirm that this description is exact
in the limit of tightly-bound wells, and we explore further to what degree this description remains accurate as the
coupling between wells increases. Remarkably, we find that the need for so-called next-nearest-neighbour (nnn)
hopping does not necessarily imply next-nearest-neighbour hopping. This exercise, carried out here only for this
specific model10, gives us a deeper understanding of the connection between effective models and their more
microscopic counterparts.
These results complement the phenomenological fits realized in refs2,7. The first reference describes a
double-well potential, and can be thought of as a special preliminary case of the fully periodic solid. This form of
the double-well potential treated in this reference is not directly applicable to the present problem but, for com-
pleteness, we include its solution separately in Appendix B.

Kronig-Penney Model
The one-dimensional Kronig-Penney model10 consists of an electron moving in a periodic potential as depicted
in Fig. 1, with alternating wells of width w and barriers of width b and height V0. The basic building block of this
model is the unit cell, here consisting of one well surrounded by two (half) barriers; in the limit that the barrier
heights and/or widths become large, a good starting point for this problem is the solution to the problem of a
single well, and we will exploit this in what follows. However, the physics that emerges from the Kronig-Penney
model is already contained in the double-well potential, whose solution is given in Appendix A (see also Appendix
B for the solution to the double-well considered in ref.2),
Proceeding with the Kronig-Penney model, the analytical solution for the energy levels (E < V0) is well known;
the implicit equation for the energy is

κ 22 − q2
cos(k ) = cos(qw ) cosh (κ 2b) + sin(qw ) sinh (κ 2b),
2 qκ 2 (1)

where  = w + b is the unit cell length, and q = 2mE / 2 and κ 2 = 2m(V0 − E )/ 2 . For each wave vector k,
with values −π < k  ≤ π , one needs to solve this equation for E(k). A similar equation holds for E > V0 (replace
κ2 → ik2, where k 2 ≡ 2m(E − V0)/ 2 ). As is well known, the periodicity in the problem gives rise to a series of
energy bands as a function of wave vector k, with each band separated by an energy gap. In the case where the
wells illustrated in Fig. 1 are deep, any single well, taken in isolation, would consist of a number of different energy
levels corresponding to states that are bound within each well. As already stated, when N of these wells are cou-
pled through barriers, each of these energy levels broadens into N states, forming bands. Numerical solutions to
this and other periodic models with different potential shapes are given in refs6 and7.

Tight-Binding
The tight-binding limit tends to focus on one of these bands, and is used to describe the dispersion, E(k) for this
band. General considerations5 in the tight-binding limit in one dimension lead to a dispersion of the form
E(k) = Eb − 2t1 cos(k ) − 2t2 cos(2k ) − 2t3 cos(3k ) − … (2)
The usual interpretation of such a dispersion is that each additional term corresponds to tunneling of an electron
from a well to a further neighbouring well. In other words, while t1 represents a tunneling amplitude for an elec-
tron to tunnel through one of the barriers in Fig. 1, t2 represents a tunneling amplitude for the electron to tunnel
through two of the barriers, and end up (directly) two unit cells away from its initial location. Given that the
electron wavefunctions are exponentially decaying in the barrier regions, it should be clear that
t1  t2  t3  … in this limit. In what follows we will first focus on nearest-neighbour tunneling ampli-
tudes only, i.e. we will obtain from Eq. (1), an explicit expression for t1.
Motivated by the case when V0 or b is suitably large, one can rewrite Eq. (1) to obtain, without approximation,

Scientific REPOrTS | 7: 17041 | DOI:10.1038/s41598-017-17223-2 2

www.nature.com/scientificreports/

 qw q qw   qw κ qw 
cos − sin  cos + 2 sin  = η1(k) + η2
 2 κ2 2   2 q 2  (3)
where

η1(k) = 2 e−κ 2b cos (k )

 qw κ qw   qw q qw 
η2 = −e−2κ 2b cos − 2 sin  cos + sin .
 2 q 2   2 κ2 2  (4)
Written in this way, it is easy to see that when there is no coupling between the wells (e.g. put b → ∞) and there-
fore both η1(k) → 0 and η2 → 0, then the vanishing of the first (second) factor on the LHS of Eq. (3) corresponds
to determining the energy for the even (odd) bound states in the single well. It is convenient to define dimension-
less variables as before, specifically z ≡ qw/2 and z0 ≡ k0w/2, where k 0 ≡ 2mV0/ 2 . Then Eq. (3) reads
  z 02 − z 2 
 z  
cos z − sin z  cos z + sin z  = η1(k) + η2,
 2
z0 − z 2   z 
  (5)

where we have used κ 2w /2 = z 02 − z 2 and now

b z 02 − z12
η1(k) = 2e−2 w cos k  (6)
and
 z 02 − z 2  
z 02 − z12   z 
sin z  cos z + sin z .
b
η2 = e−4 w cos z −
 z   2
z0 − z 2 
  (7)
Eq. (5) is still exact; now we can imagine the scenario where b/ is very large, and hence the RHS of this equation
is very small. The zeroth order solution for the even bound states is given by setting the first factor on the LHS of
Eq. (5) to zero,
z1
cos z1 − sin z1 = 0,
z 02 − z12 (8)
and this determines the zeroth order solution, z1. The equation to determine z1 can be written as
2
 z0 
tan z1 =   − 1 ,
 z 
1 (9)
which is the equation that determines the even bound state energies for a particle in a single well of width w and
depth V0. The solution is shown graphically in Fig. 2. An actual number for z1, slightly less than π/2, is readily
obtained for the lowest bound state, either numerically or on a calculator.

First-order in tunneling. A more accurate solution to Eq. (5) can be obtained to 1st order in η1(k) by writing
z = z1[1 + ρ(k)], and expanding that equation to 1st order in ρ(k). After some algebra (see Appendix C) we
obtain

2 (z 02 − z12) 2b z 02 − z12
ρ(k) = − 2
e− w cos k  .
z 0 1 + z 0 − z1
2 2
(10)
( )
Note that (for now) we have ignored η2, as that factor is exponentially suppressed with respect to η1(k). Using the
energy scale E0 = ħ2/(2mw2) as before, we find for the energy to 1st order in η1(k),
2
E  4z  z 02 − z12 z 02 − z12
= 4z12 −  1 
b
e−2 w cos k 
E0 
 z 0  1 + z 2 − z 2
0 1 (11)
which has the functional form of nearest-neighbour tight-binding (see Eq. (2)). A comparison of this result (dot-
ted blue curve) with the exact result (solid red curve) is shown in Fig. 3 as a function of wave vector for particular
values of V0 and b/.
If we first focus on case (a) in Fig. 3, the first-order result has significant disagreement with the exact result. It
is indeed true that improved agreement is readily attained by using deeper wells, as is clear from the progression
through (b–d). However, as we now illustrate, more accurate solutions are achievable by including contributions
from η2, which is already 2nd order in e−x, where x ≡ 2 b z 02 − z12 —see Eq. (7).
w

Scientific REPOrTS | 7: 17041 | DOI:10.1038/s41598-017-17223-2 3

www.nature.com/scientificreports/

Figure 2. Graphical solution of Eq. (9), for an example z0 = 2.8π. The LHS is shown with the thin solid (green)
curves, with the obvious characteristic branches of the tan function. The thicker solid (red) curve represents the
RHS. The intersections of these two curves represent the even bound states and are indicated by open squares;
the lowest energy solution is just below z1 = π /2. For completeness we have also drawn the LHS and RHS for
the odd bound states. The LHS is given by tan(z1 − π /2) and is shown with thin dashed (blue) curves. The RHS
is the same solid (red) curve as for the even states. Their intersections are indicated by filled squares. For our
tight binding solutions we will focus on the lowest energy (even) bound state, i.e. the point with z1 < π /2.


Figure 3. Comparison of various approximations with the exact result (solid (red) curve). All results are for
b/ = 0.2 and are for (a) v0 = 50, (b) v0 = 100, (c) v0 = 200, and (d) v0 = 1000. In (a), for example, the first-order
result is given by Eq. (11) (shown as the dashed (blue) curve) with the exact result determined numerically from
Eq. (1) (shown as the solid (red) curve). Note that there remains a significant discrepancy. The zeroth-order
result, a constant given by the first term only in Eq. (11), is shown as the horizontal (pink) line [e.g. at
Eb/E0 ≈ 5.94 in (a)]. With further second-order corrections that arise from η2 and the nonlinear nature of Eq.
(5), we also show the results from Eq. (16). This result now agrees very well with the exact result, and includes
terms corresponding to −2t2 cos(2k ) in Eq. (2). Notice that for the last case, all curves and points are essentially
in agreement. Given the reduction in the scale of the bandwidth in (a) → (d), this is impressive agreement.

Scientific REPOrTS | 7: 17041 | DOI:10.1038/s41598-017-17223-2 4

www.nature.com/scientificreports/

Second-order in tunneling. We first repeat the steps leading to Eq. (10), i.e. we write the solution as
z ≈ z1(1 + ρ(k)), but now expand to 2nd order in ρ(k) as well as in e−x, since η2/η1(k) ≈ O(e−x). A straightforward
calculation gives
2
z0 2b δ 2
−ρ(k) {1 − ρ(k)g1} = 2e−x cos k  + 2z 0 e−xρ(k) cos k  + 2e−2x(1 − 2δ ),
f1 w 1 − δ
2
(12)
where
2
1 − δ
f1 =
2 1
1 − δ + z0
2 3
δ
2 1 − δ + 2z 0
g1 = 1 − 2
,
1 − δ 2
1 − δ +
1
z0 (13)
and δ ≡ z1/z 0. Eq. (12) is a quadratic in ρ(k); since this is a small quantity we can simply iterate to obtain ρ(k)
explicitly:
2 −x
ρ(k) = − f e cos k 
z0 1
 f  2 
2f  2 2b δ 
− 1 e−2x 1 − 2δ − 1 g1 + z 0 
z0  z  w 2 
 0  1 − δ 


2f 2  2 
 2b δ 
+ 12 e−2x g1 + z 0  cos2k  .
 w
z0 1 − δ 
2
 (14)
Finally, we use

E(k) = 4E 0z12(1 + ρ)2 (15)

from which we obtain
E(k) = Ec − 2t1 cos k  − 2t2 cos2k  , (16)
where the parameters defined by Eq. (16) are given by
2
t1 = 8E 0z 0δ f1 e−x , (17)

 2 
2  1 2b δ  −2x
t2 = − 8E 0δ f12 g1 + + z 0 e

 2 w 1 − δ 
2
(18)
and
2
Ec = E 0{4z12 − 16z1δf1 e−2x(1 − 2δ )} − 2t2. (19)
Note that the expansion is governed by the exponential suppression contained in the e−x and e−2x factors.
However, we expect z0 > 1 for tight-binding, whereas Fig. 2 makes it clear that z1 is of order unity or lower; more
precisely, z1 < min(z 0, π /2), and so δ < 1 as well. We have written the above expressions to make these expan-
sion parameters clear; thus, even within the expression for t2, for example, various terms will contribute signifi-
cantly less than others. Both f1 and g1 are of order unity.

Discussion
The more accurate result from Eq. (16) is plotted in Fig. 3 (as indicated by the square symbols), and gives good
agreement for the parameters used in that figure. Note that t1 is given by an expression identical to the one implied
in Eq. (11)—by going to 2nd order in e−x this has not changed, and in fact is identical to the expression derived for
the double-well in Appendix A, Eq. (A5). The need to to go to 2nd order and therefore generate a term with
cos(2k ) wave vector dependence is sometimes interpreted to mean that a significant tunneling amplitude exists
between second-nearest-neighbour atoms. The fact that this is required even for the case of the double-well stud-
ied in Appendix A (where there is no second-nearest neighbour!) indicates that this interpretation is incorrect.
Indeed it is a difficult problem to disentangle contributions to 2nd order from next-nearest-neighbour tunneling
and contributions arising from the inherent non-linear nature of the equations; at this point we simply caution
that all these contributions are not entirely due to direct next-nearest-neighbour tunneling.
For completeness, in Fig. 4 we fix the well depth to be v0 ≡ V0/E0 = 50 and show the dispersions for a variety of
different barrier widths. The trends are the same in the two cases, except that the 2nd order result is slightly less

Scientific REPOrTS | 7: 17041 | DOI:10.1038/s41598-017-17223-2 5

www.nature.com/scientificreports/

Figure 4. As in Fig. 3, a comparison of various approximations with the exact result (solid (red) curve) for a
variety of barrier widths, all with v0 ≡ V0/E0 = 50. We use barrier widths of (a) b/ = 0.1, (b) b/ = 0.2, (c)
b/ = 0.3 and (d) b/ = 0.4. By examining the vertical scales, a clear progression towards more tightly bound
wells is evident. Moreover, the first case considered shows not perfect agreement, even when 2nd order
corrections are included. As b/ increases, agreement quickly improves, especially when account is made of the
steady reduction in the scale of the bandwidth. For all cases, the horizontal (pink) line corresponds to the bound
state energy for a single well and corresponds to the constant given by the first term only in Eq. (11). The
approximate result with the 1st order correction only, the full Eq. (11), is shown as a dashed (blue) curve, while
the approximate results including 2nd order corrections from Eq. (16) are given by the square (green) symbols.

accurate for the least tightly bound case considered [see (a)]. With increasing barrier width, however, as in Fig. 3,
both the 2nd-order and the 1st-order results become increasingly accurate. Note the change in scale as v0 and b/
increase in Figs 3 and 4, respectively; in both cases the results approach the single well result while a well-defined
dispersion remains.
In both Figs 3 and 4 it should be clear that for the more strongly coupled wells (e.g. (a) and (b) in particular)
a significant amount of electron-hole asymmetry is present. In Fig. 5 the effective mass ratio, |mh/me| is shown
for various values of the barrier thickness as a function of well depth. Here, the electron mass, me is defined in
the usual way (see the caption in Fig. 5) through the curvature at k = 0 and similarly for the hole mass, mh. As
discussed in ref.6 an asymmetry is expected on general grounds since holes are by definition closer to the top of
the barriers than electrons. They should therefore have lower masses for this reason alone, and this is reflected in
the results of Fig. 5, where all the ratios are lower than unity. The thicker curves are from the exact calculations
while the thinner curves (in too good agreement with the exact results to be visible for most of the parameter
space shown) are readily determined from the tight-binding parametrization of Eq. (16). These are fairly accurate
when the higher-order correction considered above is included.
Figure 5 exemplifies what is perhaps the primary result of this calculation—that electron-hole asymmetry is
the “rule”, not the exception. In the past three decades, as more materials with exotic properties (superconductiv-
ity, magnetoresistance, etc.) have been engineered, particularly through doping, and electron-hole-doping sym-
metry has not been observed, strong correlations are often suggested as a possible explanation. The present work
notes that the usual starting point, tight-binding with nearest-neighbour hopping only, tends to bias the expecta-
tion in favour of electron-hole symmetry, when in fact the opposite is true. Our calculations illustrate that even
where merely nearest-neighbour hopping seems like a good physically-motivated assumption, the band structure
(before invoking interaction terms) will be already electron-hole asymmetric. For example, Hirsch and one of the
present authors11 suggested a modulated hopping model in which one of the interactions is explicitly particle-hole
asymmetric in a tight-binding framework (see ref.12 for an overview of the physics of hole superconductivity). It

Scientific REPOrTS | 7: 17041 | DOI:10.1038/s41598-017-17223-2 6

www.nature.com/scientificreports/

Figure 5. Absolute value of the effective mass ratio, |mh/me| vs. v0 ≡ V0/E0 for the various values of barrier width
2 2
b/, as indicated. Here me−1 ≡ ∂ E(k) /2E 0 at k = 0 and mh−1 ≡ ∂ E(k) /2E 0 at k = π /. The thick curves are the exact
∂(k ) ∂(k )
result, determined from Eq. (1), while the thinner curves are determined from the tight-binding
parametrization of Eq. (16). These latter curves are barely visible over almost the entire parameter regime
shown, indicating that when the 2nd order corrections considered are included the derived tight-binding
parameters are very accurate.

follows that many of the properties of this model will exhibit electron-hole asymmetries. An accurate confron-
tation with experiment will required some knowledge of the asymmetry already possibly present in the band
structure, not necessarily because of a next-nearest-neighbour hopping term, but for the reasons suggested in this
paper—the smaller tunneling barrier faced by holes compared to electrons.

Summary
We have succeeded in deriving the effective model for the periodic potential first used to model a solid, the
so-called Kronig-Penney model, consisting of a series of wells and barriers. As is more readily illustrated for the
double-well potential, one can achieve very high accuracy by exploiting the tightly-bound limit, where two neigh-
bouring wells are well separated. This ensures that the tunneling amplitude between the two wells is very small,
and one can essentially use perturbation theory with respect to this “atomic limit”.
The generalization of this process to an infinite array of wells and barriers is straightforward. However, incor-
porating tunneling to 1st order (meaning terms of order e−x, where x ∝ b V0 ) provided only a qualitative agree-
ment with the exact result (this would become quantitative for sufficiently large b/ and/or V0). When including
terms of 2nd order in e−x, very good quantitative agreement was achieved, even for moderate well depths. Terms
of order e−2x would necessarily be accompanied by dispersive terms like cos(2k ), which are generally associated
with next-nearest-neighbour tunneling, i.e. tunneling across two barriers. By comparison with results of a simple
double-well, where terms of order e−2x also contribute to the energy, we were able to show that terms of this order
were not exclusively associated with such longer-range tunneling. Instead, inherent non-linearity of the equations
governing the electronic energy dispersion will naturally give rise to such terms, even in the absence of
next-nearest-neighbour tunneling.
It is useful to map the complete microscopic double-well problem onto the two-state system that is often used
to describe this problem in simplified terms. Similarly, it is useful to map the microscopic problem of an infinite
array of wells onto a simplified model—this is the tight-binding description. We have carried out such a mapping,
with no “fitting” involved and we have illustrated the accuracy as well as the limitations of such a mapping.

References
1. Feynman, R. P., Leighton, R. B. & Sands, M. The Feynman Lectures on Physics, Volume III (Addison-Wesley, Reading, MA, 1965).
2. Dauphinee, T. & Marsiglio, F. Asymmetric wave functions from tiny perturbations. Am. J. Phys. 83, 861–866 (2015).
3. Jelic, V. & Marsiglio, F. The double-well potential in quantum mechanics: a simple, numerically exact formulation. Eur. J. Phys. 33,
1651–1666 (2012).
4. Bloch, F. Über die Quantenmechanik der Elektronen in Kristallgittern. Z. Phys. 52, 545–555, in German (1929).
5. Ashcroft, N. W. & Mermin, N. D. Solid State Physics. 1st ed., See Appendix E. (Brooks/Cole, Belmont, CA, 1976).
6. Pavelich, R. L. & Marsiglio, F. The Kronig-Penney model extended to arbitrary potentials via numerical matrix mechanics. Am. J.
Phys. 83, 773–781 (2015).
7. Pavelich, R. L. & Marsiglio, F. Calculation of 2D electronic band structure using matrix mechanics. Am. J. Phys. 84, 924–935 (2016).
8. Bloch, F. Memories of electrons in crystals. Proc. R. Soc. Lond. A 371, 24–27 (1980).
9. See, for example, Nomura, Yusuke Ab initio studies on superconductivity in alkali-doped fullerides. (Springer, Toronto, 2014).
10. Kronig, R. deL. & Penney, W. G. Quantum Mechanics of Electrons in Crystal Lattices. Proc. R. Soc. Lond. A 130, 499–513 (1931).
11. Hirsch, J. E. & Marsiglio, F. Superconducting state in an oxygen hole metal. Phys. Rev. B 39, 11515–11525 (1989).
12. Hirsch, J. E. Towards an understanding of hole superconductivity. High-Tc Copper Oxide Superconductors and Related Novel
Materials, Vol. 255, 99–115, edited by Bussmann-Holder, Annette, Keller, Hugo & Bianconi, Antonio (2017).

Scientific REPOrTS | 7: 17041 | DOI:10.1038/s41598-017-17223-2 7

www.nature.com/scientificreports/

Acknowledgements
This work was supported in part by the Natural Sciences and Engineering Research Council of Canada (NSERC)
and originated in work originally funded by a University of Alberta Teaching and Learning Enhancement Fund
(TLEF) grant, for which we are grateful.

Author Contributions
F.M. originally formulated the problem addressed in this paper. Both authors contributed to the calculations and
writing of the manuscript.

Additional Information
Supplementary information accompanies this paper at https://doi.org/10.1038/s41598-017-17223-2.
Competing Interests: The authors declare that they have no competing interests.
Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International
License, which permits use, sharing, adaptation, distribution and reproduction in any medium or
format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Cre-
ative Commons license, and indicate if changes were made. The images or other third party material in this
article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the
material. If material is not included in the article’s Creative Commons license and your intended use is not per-
mitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the
copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.

© The Author(s) 2017

Scientific REPOrTS | 7: 17041 | DOI:10.1038/s41598-017-17223-2 8